RAG Development
Comprehensive knowledge base for building production-grade Retrieval-Augmented Generation systems.
When to Use
- Building a new RAG pipeline from scratch
- Choosing chunking strategy, embedding model, or vector database
- Implementing hybrid search, re-ranking, or contextual retrieval
- Evaluating RAG quality with RAGAS or DeepEval
- Optimizing production RAG for cost, latency, or accuracy
- Designing multi-tenant RAG with access control
- Upgrading from naive RAG to advanced patterns
Quick Start Recommendation
For 80% of use cases, start with:
- Chunking: Recursive character splitting at 512 tokens, 10-15% overlap
- Embedding: OpenAI
text-embedding-3-small (best value) or Cohere embed-v4 (best accuracy)
- Vector DB: Qdrant with scalar INT8 quantization
- Retrieval: Hybrid search (dense + sparse + RRF)
- Evaluation: RAGAS metrics from day one
Then upgrade incrementally based on measured failures:
- Keyword misses -> add sparse vectors (SPLADE/BM25)
- Ambiguous chunks -> add contextual retrieval (Anthropic pattern)
- Irrelevant results -> add cross-encoder re-ranking
- Multi-hop failures -> upgrade to agentic RAG
Reference Materials
Detailed reference documents are in the references/ directory:
chunking-strategies.md -- all chunking approaches with code, benchmarks, and selection guideembedding-models.md -- model comparison, Matryoshka embeddings, fine-tuning, sparse/dense/multi-vectorretrieval-patterns.md -- hybrid search, HyDE, contextual retrieval, re-ranking, MMRadvanced-rag-patterns.md -- Graph RAG, RAPTOR, CRAG, Self-RAG, Agentic RAG, multi-modal RAGvector-databases.md -- Qdrant deep dive, database comparison, scaling strategiesproduction-guide.md -- evaluation, observability, caching, security, cost optimization
Pipeline Architecture
Document Ingestion:
Raw Docs -> Preprocessing (Unstructured.io) -> Chunking -> Context Enrichment -> Embedding -> Vector DB
Query Pipeline:
User Query -> Query Transform -> Encode (Dense + Sparse) -> Hybrid Search -> Re-rank -> LLM Generation
Evaluation Loop:
Ground Truth + Predictions -> RAGAS/DeepEval -> Faithfulness, Relevancy, Precision, Recall
Key Decision Points
| Decision |
Default |
Upgrade When |
| Chunking |
Recursive 512 tok |
Structured docs -> markdown-aware; cross-refs -> late chunking |
| Embedding |
text-embedding-3-small |
Need accuracy -> embed-v4; self-hosted -> NV-Embed-v2 |
| Vector DB |
Qdrant + INT8 |
Already on Postgres -> pgvector; need managed -> Pinecone |
| Search |
Dense only |
Keyword misses -> add sparse hybrid; poor diversity -> add MMR |
| Re-ranking |
None |
Top-k results contain irrelevant items -> add Cohere Rerank |
| Caching |
None |
Production latency/cost concerns -> semantic cache |
| Evaluation |
Manual spot checks |
Any production use -> RAGAS automated metrics |
1---2name: rag-development3description: Knowledge base for production question-answering pipelines. TRIGGER WHEN: building, optimizing, or auditing RAG systems, including Graph RAG, CRAG, Self-RAG, and Agentic RAG.4---56# RAG Development78Comprehensive knowledge base for building production-grade Retrieval-Augmented Generation systems.910## When to Use1112- Building a new RAG pipeline from scratch13- Choosing chunking strategy, embedding model, or vector database14- Implementing hybrid search, re-ranking, or contextual retrieval15- Evaluating RAG quality with RAGAS or DeepEval16- Optimizing production RAG for cost, latency, or accuracy17- Designing multi-tenant RAG with access control18- Upgrading from naive RAG to advanced patterns1920## Quick Start Recommendation2122For 80% of use cases, start with:231. **Chunking**: Recursive character splitting at 512 tokens, 10-15% overlap242. **Embedding**: OpenAI `text-embedding-3-small` (best value) or Cohere `embed-v4` (best accuracy)253. **Vector DB**: Qdrant with scalar INT8 quantization264. **Retrieval**: Hybrid search (dense + sparse + RRF)275. **Evaluation**: RAGAS metrics from day one2829Then upgrade incrementally based on measured failures:30- Keyword misses -> add sparse vectors (SPLADE/BM25)31- Ambiguous chunks -> add contextual retrieval (Anthropic pattern)32- Irrelevant results -> add cross-encoder re-ranking33- Multi-hop failures -> upgrade to agentic RAG3435## Reference Materials3637Detailed reference documents are in the `references/` directory:3839- `chunking-strategies.md` -- all chunking approaches with code, benchmarks, and selection guide40- `embedding-models.md` -- model comparison, Matryoshka embeddings, fine-tuning, sparse/dense/multi-vector41- `retrieval-patterns.md` -- hybrid search, HyDE, contextual retrieval, re-ranking, MMR42- `advanced-rag-patterns.md` -- Graph RAG, RAPTOR, CRAG, Self-RAG, Agentic RAG, multi-modal RAG43- `vector-databases.md` -- Qdrant deep dive, database comparison, scaling strategies44- `production-guide.md` -- evaluation, observability, caching, security, cost optimization4546## Pipeline Architecture4748```49Document Ingestion:50 Raw Docs -> Preprocessing (Unstructured.io) -> Chunking -> Context Enrichment -> Embedding -> Vector DB5152Query Pipeline:53 User Query -> Query Transform -> Encode (Dense + Sparse) -> Hybrid Search -> Re-rank -> LLM Generation5455Evaluation Loop:56 Ground Truth + Predictions -> RAGAS/DeepEval -> Faithfulness, Relevancy, Precision, Recall57```5859## Key Decision Points6061| Decision | Default | Upgrade When |62|----------|---------|-------------|63| Chunking | Recursive 512 tok | Structured docs -> markdown-aware; cross-refs -> late chunking |64| Embedding | text-embedding-3-small | Need accuracy -> embed-v4; self-hosted -> NV-Embed-v2 |65| Vector DB | Qdrant + INT8 | Already on Postgres -> pgvector; need managed -> Pinecone |66| Search | Dense only | Keyword misses -> add sparse hybrid; poor diversity -> add MMR |67| Re-ranking | None | Top-k results contain irrelevant items -> add Cohere Rerank |68| Caching | None | Production latency/cost concerns -> semantic cache |69| Evaluation | Manual spot checks | Any production use -> RAGAS automated metrics |
acaprino